"Hasn't" given any intuitive reading of the term `utility', that is. One could say that impossible goals can be modelled by either attributing it a zero utility or an infinite cost. However this abstracts the meaning of the term away from any conception of usefulness and changes the single utility optimization model into more of an abstract "black-box" model. In this form it would cease to correspond to how agents actually make decisions - in which case there would be no reason not to attribute multiple incommensurable utilities if this was convenient.